SKILL.md

Using gpt4free as an LLM Server for Bots (Clawbot/OpenClaw)

Overview

This skill covers running gpt4free as a local LLM server with an OpenAI-compatible REST API, custom model routing (config.yaml), and integration with bots like Clawbot or OpenClaw.

Best Practices

• Start the API server with: python -m g4f --port 8080 (or use g4f api --debug --port 8080)
• Use the /v1 endpoint for OpenAI-compatible requests (e.g., POST to http://localhost:8080/v1/chat/completions)
• Define custom model routes in config.yaml to aggregate/fallback across providers
• Place config.yaml in your cookies directory (e.g., ~/.g4f/cookies/config.yaml)
• For Clawbot/OpenClaw, patch their config to point to your gpt4free server (see patch-openclaw.py)
• Test with: g4f client "Hello" --model openclaw or Python client

Common Pitfalls

• Not starting the server before connecting bots
• Incorrect config.yaml path or syntax errors
• Missing required Python dependencies (install with pip install -r requirements.txt)
• Not exposing the correct port (default 8080)
• Forgetting to patch bot configs to use your local endpoint

Workflow Steps

1. Install and set up gpt4free (see README)
2. Start the API server: python -m g4f --port 8080
3. (Optional) Create or edit config.yaml for custom model routing:

   models:
       - name: "openclaw"
           providers:
               - provider: "GeminiCLI"
                   model: "gemini-3-flash-preview"
                   condition: "quota.models.gemini-3-flash-preview.remainingFraction > 0 and error_count < 3"
               - provider: "Antigravity"
                   model: "gemini-3-flash"
               - provider: "PollinationsAI"
                   model: "openai"
   

4. Patch your bot config (e.g., OpenClaw) to use http://localhost:8080/v1 as the base URL (see scripts/patch-openclaw.py)
5. Start your bot and verify it connects to gpt4free
6. Monitor logs and test with the Python client or CLI

References

• README.md
• docs/config-yaml-routing.md
• scripts/patch-openclaw.py
• scripts/setup-openclaw.sh
• g4f/client/init.py